BENGALURU: As companies race to deploy GenAI, a new challenge is emerging for IT companies: token maximization. Tokens — units that are consumed when large language models process information — have become the primary currency of artificial intelligence.
But after an initial wave of experiments, IT services companies are increasingly warning that relentless token consumption without measurable business results could become the industry’s next cost problem.Token maxxing treats AI consumption as a measure of productivity, driving employees and organizations to use increasing amounts of computing and tokens. The term has gained currency in Silicon Valley, where it has more than doubled in size “xx” in “com. maxxing”Borrowed from Internet and gaming slang, it refers to aggressive optimization.
As organizations move from pilots to large-scale deployments, IT companies are increasingly focusing on linking token use to business value rather than raw consumption.
Arumugam Kumaradasan, Vice President and Head of Manufacturing AI and IT Automation at Cognizant, said: “Tokens are an input to delivery, not a measure of value, and token consumption is merely a cost signal, tracked for discipline, license management and capacity planning.
When token consumption is treated as the primary metric, costs scale linearly with demand without a corresponding return in business outcomes.”
Kumaradasan said the company has implemented measurement and value mapping capabilities that link token consumption to business workflows and outcomes. “As the industry moves toward results-based models, nominal spending will reveal the cost of achieving those results,” He said.At Happiest Minds, Executive Vice President Joseph Anantharaju said the company is developing capabilities to scale and optimize the token as organizations scale AI deployments. “I think that’s going to be very important – being able to measure it,” He said.The company is also evaluating outcomes-based business models that combine software, agents, platforms and AI consumption.“The results customers want can be achieved through a model that combines the solution, token consumption and number of agents deployed,” Anantharaju said.This debate is becoming increasingly important as AI projects move from experimental to production. “One of the most important currencies in agent AI conversion are tokens. We now look at our symbolic consumption,” said Vala Afshar, chief digital officer at Salesforce. “We’re also looking closely at how much of these tokens are actually automating tasks by agents, because it’s wasteful to just spend tokens unless you’re creating value as quickly as needed.
” However, technology providers have become wary of turning token consumption into a pricing model.Nitin Rakesh, CEO of Mphasis, said customers are increasingly incurring variable nominal costs as usage expands. “What we price is the economic outcome. There is a base price that you (the client) will pay me, and the rest will be tied to the result I can achieve,” He said.